The most common purpose of seasonal adjustment is to provide an estimate of the current trend so that judgmental short-term forecasts can be made. Bell (Proceedings of the American Statistical Association, 1995) formally considered how model-based seasonal adjustment could be done in order to facilitate the forecasting, showing that, from a theoretical perspective, this objective could be best served by not revising the data for seasonality. However, this study was lacking an empirical investigation to determine if this approach would realize any advantages when applied in practice. Aim of this paper is to assess whether is convenient, from forecasting perspective, to adjust data for seasonality or directly use a model which accounts for seasonality. In particular, we serve this scope by both a simulation study and an application with real data related to Industrial Production Index. We show that pre-adjusting the time series for seasonality allows for forecasting improvements in terms of accuracy. In the end we evaluated also the best seasonally adjustment method for forecasting purposes. Empirical evidence shows that the forecasts among seasonal adjustment methods are statistically different and that, while for the Italian TRAMO-SEATS outperform X13-ARIMA-SEATS, the opposite happen for the U.S. Industrial Production. Results from the simulation suggest that the forecasts among the considered seasonal adjustment methods are not statistically different for very short and long time period. However, seasonal adjustment methods lead to statistically different forecasts for medium and for very long time period.